GCM Help Pages
Main Before you start Computing GCMs Viewing GCMs Interpreting GCMs Advanced More Help	Advanced The GCMPlugin reads some of parameters from the 'GCMPlugin.ini' text-file at start-up, which should be next to the GCMPlugin dll. These parameters are 'Global' parameters, in the sense that they generally affect all computations of every map. Most the parameters fine-tune the computations of the Granger Causality Maps, their various compositions (e.g. dGCMs) and the bootstrapping of significance thresholds. Some parameters can be set to taste, such as: which result maps and which FDR thresholds to compute, how much feedback to give in the Log-tab, and how to scale the maps for visualisation. Most computation related parameters have generally been assigned sensible or even optimal defaults, but you can change them to see what effect a different setting might have. The .ini file is anotated with comment-lines (starting with '//') that explain the different parameters. If an unacceptable change is made to the .ini file, the GCMPlugin will give out a warning at start-up, generally with a remark about the offending parameter. The GCMPlugin will use the default parameters, hard-coded into GCMPlugin.dll, and continue when the .ini file could not be found or was invalid. Therefore, be attentive to warnings at start-up to see whether your changes did not invalidate the .ini file and were subsequently ignored. //***************************************************** // Granger causality mapping (GCM) plugin ini-file // Read by the GCM-Plugin to initialize global // parameters. Change only if you know what // you are doing and at your own risk! //***************************************************** // Maps that are computed for every condition, possible values (enumeration): // INST : Instanteneous GC term between reference and voxel (Geweke (1982)) // REF2VOX : Directed GC term from reference to voxel (Geweke (1982)) // VOX2REF : Directed GC term from voxel to reference (Geweke (1982)) // DGCM : The difference GCM (DGCM) = REF2VOX - VOX2REF // DGCM_IT : The dGCM, instantaneously thresholded, i.e. dGCM masked with // the thresholded instantaneous map, // i.e. DGCM_IT = DGCM, only where INST > DGCM_ITInstLowerThres (unscaled, see below) // else it is set to zero (vessels) MapTypesComputed= INST DGCM_IT DGCM_ITInstLowerThres= 0.05 // False dicovery rates (q) used to compute (multiple comparison corrected) thresholds // Listed in the map dialog QThresholdsForFDR= 0.01 0.05 0.1 // The c(V) constant (see Genovese, NI) that incorporates assumptions about the spatial // distribution of p-values over the map. Can be 1 (c(V)=1, liberal, some assumptions), // or 2 (c(V)=sum(1:N)), conservative, but assumption free) CVConstantForFDR= 2 // Show a verbose progress-log in the Log-tab while computing, when 1 // Only show limited information (e.g. written files), when 0 ShowVerboseProgressLog= 1 // All maps (except DGCM_RC that always ranges from 0 to 1) are scaled by the MapScaleFactor // This is only for convenience and visualisation purposes. The natural values of the computed // measures lie in the 0.01 range. The scale-factor may be used to put them in a more // 'intuitive' range, that works beter with display and thresholding routines in BVQX. MapScaleFactor= 100 // Order of the VAR models used to compute the Geweke (1982) measures AutoRegressionOrder= 1 // Voxels with first time-point below FirstTimePointInclusionThres are discarded, // All GCMs are set to zero, and are not taken into account for threshold determinations FirstTimePointInclusionThres= 100 // Bootstrapping of p-values is done by finding the fraction of more extreme values // in a surrogate NULL-distribution. The NULL-distribution is created by recomputing each // term for each voxel with a simulated NULL-reference (from an AR model estimated on // the real reference). Instead of resimulating a reference for each voxel, a simulated // reference is randomly selected for each voxel from a pre-simulated pool of NULL-refs. // Pool has size NumSurrogateSimulations NumSurrogateSimulations= 5000 // When simulating an AR model (driven by white noise), the first part of the simulated series // has to be discarded, to get to a steady state of the model, NumDiscardedSimulatedTimePoints points // are skipped for every simulation in the pre-simulated pool of NULL-refs NumDiscardedSimulatedTimePoints= 100

GCM Help Pages

Main
Before you start
Computing GCMs
Viewing GCMs
Interpreting GCMs
Advanced
More Help

Advanced

The GCMPlugin reads some of parameters from the 'GCMPlugin.ini' text-file at start-up, which should be next to the GCMPlugin dll. These parameters are 'Global' parameters, in the sense that they generally affect all computations of every map. Most the parameters fine-tune the computations of the Granger Causality Maps, their various compositions (e.g. dGCMs) and the bootstrapping of significance thresholds. Some parameters can be set to taste, such as: which result maps and which FDR thresholds to compute, how much feedback to give in the Log-tab, and how to scale the maps for visualisation. Most computation related parameters have generally been assigned sensible or even optimal defaults, but you can change them to see what effect a different setting might have. The .ini file is anotated with comment-lines (starting with '//') that explain the different parameters.

If an unacceptable change is made to the .ini file, the GCMPlugin will give out a warning at start-up, generally with a remark about the offending parameter. The GCMPlugin will use the default parameters, hard-coded into GCMPlugin.dll, and continue when the .ini file could not be found or was invalid. Therefore, be attentive to warnings at start-up to see whether your changes did not invalidate the .ini file and were subsequently ignored.

//*******************************************************
// Granger causality mapping (GCM) plugin ini-file
// Read by the GCM-Plugin to initialize global
// parameters. Change only if you know what
// you are doing and at your own risk!
//*******************************************************

// Maps that are computed for every condition, possible values (enumeration):
// INST : Instanteneous GC term between reference and voxel (Geweke (1982))
// REF2VOX : Directed GC term from reference to voxel (Geweke (1982))
// VOX2REF : Directed GC term from voxel to reference (Geweke (1982))
// DGCM : The difference GCM (DGCM) = REF2VOX - VOX2REF
// DGCM_IT : The dGCM, instantaneously thresholded, i.e. dGCM masked with
// the thresholded instantaneous map,
// i.e. DGCM_IT = DGCM, only where INST > DGCM_ITInstLowerThres (unscaled, see below)
// else it is set to zero (vessels)
MapTypesComputed= INST DGCM_IT
DGCM_ITInstLowerThres= 0.05

// False dicovery rates (q) used to compute (multiple comparison corrected) thresholds
// Listed in the map dialog
QThresholdsForFDR= 0.01 0.05 0.1
// The c(V) constant (see Genovese, NI) that incorporates assumptions about the spatial
// distribution of p-values over the map. Can be 1 (c(V)=1, liberal, some assumptions),
// or 2 (c(V)=sum(1:N)), conservative, but assumption free)
CVConstantForFDR= 2

// Show a verbose progress-log in the Log-tab while computing, when 1
// Only show limited information (e.g. written files), when 0
ShowVerboseProgressLog= 1

// All maps (except DGCM_RC that always ranges from 0 to 1) are scaled by the MapScaleFactor
// This is only for convenience and visualisation purposes. The natural values of the computed
// measures lie in the 0.01 range. The scale-factor may be used to put them in a more
// 'intuitive' range, that works beter with display and thresholding routines in BVQX.
MapScaleFactor= 100

// Order of the VAR models used to compute the Geweke (1982) measures
AutoRegressionOrder= 1

// Voxels with first time-point below FirstTimePointInclusionThres are discarded,
// All GCMs are set to zero, and are not taken into account for threshold determinations
FirstTimePointInclusionThres= 100

// Bootstrapping of p-values is done by finding the fraction of more extreme values
// in a surrogate NULL-distribution. The NULL-distribution is created by recomputing each
// term for each voxel with a simulated NULL-reference (from an AR model estimated on
// the real reference). Instead of resimulating a reference for each voxel, a simulated
// reference is randomly selected for each voxel from a pre-simulated pool of NULL-refs.
// Pool has size NumSurrogateSimulations
NumSurrogateSimulations= 5000
// When simulating an AR model (driven by white noise), the first part of the simulated series
// has to be discarded, to get to a steady state of the model, NumDiscardedSimulatedTimePoints points
// are skipped for every simulation in the pre-simulated pool of NULL-refs
NumDiscardedSimulatedTimePoints= 100